gentoo, if you've gone to that much trouble to minimize input lag, you're going to like some new features in the upcoming WA 3.8.0.
There is a DirectDraw 8-bit (hardware) mode, which on my Windows XP installation yields framerates upward of 900 fps with a normal (1920×696) map. I have found this to have lower input lag than anything else. I don't use it though, because it makes minimization and unminimization slow. I use Direct3D 9 (shader).
Assisted v-sync now has some configurable parameters (via the registry). Once fine-tuned correctly, it works quite well with vsync disabled, with "vsync off + assisted vsync on" yielding smooth updating with less input lag than "vsync on + assisted vsync off".
And... there is a really big new feature, which among other things, reduces effective input lag in all modes. I'll give a hint: It reduces average input lag by 10 milliseconds.
AFAIK there is no reason to use Direct3D 9 (CPU) unless Direct3D9 (shader) either doesn't work on a particular GPU or has glitches, neither of which has ever been reported. (This applies both to 3.7.2.1 and the upcoming 3.8.0.)
When I did a test with WA a while back using my video camera in 240 fps slowmo, with D3D9 (shader), I found keypresses to have about 42 ms of lag (from pounding a key fast/hard to seeing the effect on-screen), frontend hardware mouse cursor to have about 27 ms of lag (from suddenly shoving the mouse to seeing the effect on-screen), and in-game mouse cursor to have about 40 ms of lag. This is with an Northgate OmniKey Ultra2 keyboard plugged into the PS/2 port, and USB mouse.
CPU rendering resulted in smoother gameplay on an older computer with high DPC latency caused by Nvidia drivers. The difference shouldn't be noticeable on most systems.
That is very interesting. I had never heard of this before.